Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Aquarium / IDeA running out of tasks?
- - By KenB Date 2020-05-29 17:05
Hi all,

New guy here.  Honestly, been lurking here for a while and learning as much as I can from this forum, which has been a treasure trove of helpful advice.  So, before I even ask my question, know that I'm incredibly grateful for everything that's been shared here on this forum over the years.

I am using AQ2020.  I have used different versions of Aquarium for many years, so I'm not new to the software.  I'm having something strange happen all of a sudden, and I'm not sure what's going on.  I started a new IDeA project for an ICCF game I'm playing.  I seeded the project with games from a database and turned on auto tree expansion.  I set it to running overnight, and when I got up this morning it wasn't running.  It had processed all of the tasks it started with, plus about 1500 more, but it had stopped.  I thought that perhaps I didn't have the "Automatic Tree Expansion" turned on, but I checked and I did.

Nonetheless, figuring I screwed something up somewhere, I tweaked all of the IDeA settings for the project again, gave it some more tasks with even more positions from a database and set it to running.  I kept an eye on it for about an hour, and initially it looked like everything was running as normal.  It processed the new tasks I gave it, it went through it's typical search for alternatives, prolongation of promising lines, etc., and kept adding new tasks and processing them.  So, I figured all was well and that last night I just had something set wrong.  Then, all of a sudden, it just ran out of tasks and stopped.  (By the way, I'm not using a custom task, which I would expect to stop once it reached the end of the prolongation strategy.)

I don't remember this ever happening.  Usually it just keeps generating new tasks and cruising along until I manually stop it.  Does anyone have any suggestions as to why this might be happening, something I should check, or info you'd need to help?  Thanks!
Parent - - By centipawn (*) Date 2020-05-29 17:37
The two settings coming to my mind are "Limit variation length" and "Project score bounds" - have you checked those?

Others may have more suggestions... I have not used it for as long as you have.
Parent - - By KenB Date 2020-05-29 18:29
I do have Limit Variation Length checked and it's set at 75.  I have the project score bounds set to -50 to 50, and I do not have the relative to root score box checked.

From the root node, if I follow the best line in the tree, it's running out to about 20 plies.  This is an interesting point you make, though, because in the past I've used wider score bounds and the expansion never stopped (at least in the time I let it run).  It had never occurred to me that these bounds would imply a stopping rule for the tree expansion.  I always assumed the bounds only limited the number of alternatives I'd see at any ply.  Now that you mention this, though, I guess it should have been obvious to me that the expansion is going to stop looking at alternatives that are outside the bounds and only travel down the paths that stay within the boundary.  If the tree expands itself to a point where every move besides the best one lies outside the set boundary, then I guess the expansion would stop. Is that correct?  I guess I had (perhaps incorrectly) assumed it would always find something to continue to investigate.
Parent - - By centipawn (*) Date 2020-05-29 18:37
I believe this is correct, yes. 50 is just half a pawn, and especially if your root (or all roots, if you have several) has an evaluation that is not close to 0, Aquarium may stop finding moves that are within +/- half a pawn. I generally check the relative to root box.

Having said this, I have also seen Aquarium add moves to the tree that have fallen outside my configured score interval, but it never seemed to go down these variants to any great depth.
Parent - - By KenB Date 2020-05-29 18:56
I was able to confirm this was this issue.  I simply went in and changed the score bounds, and it immediately added new tasks.  Thanks very much for your quick, helpful reply!
Parent - - By ChiefPushesWood (**) Date 2020-05-29 20:20
To reiterate what Centipawn said, I typically WANT 'relative to root score' checked... Once your project root score is outside the bounds parameters, it will stop producing tasks.

Parent - - By KenB Date 2020-05-29 23:16
Thanks Chief!  I've been an Aquarium user for years because I thought it was the nicest of the available GUIs.  IDeA, though, was something I only dabbled with.  However, as I'm getting older and losing interest in playing over-the-board, I've recently joined ICCF and am hoping to pursue that as my chess outlet.  As a result, I'm starting to think more about the options in Aquarium, and it's becoming clear fairly quickly that there's a lot to pay attention to to get what you want.  That's why I'm thankful for this forum and the advice and wisdom that's been shared over the years.  Even little checkboxes like this, that are likely obvious to veterans, can seem unimportant to a novice like me.  So, I appreciate the advice you and Centipawn shared on its usefulness!

Parent - - By pawnslinger (****) Date 2020-05-30 05:58
In general, one should use a score bound that is in relation to the size of the tree.  For example, on my Najdorf, which is quite complex and large (> 3 million nodes), I routinely use +- 0.18 up to +-0.75, depending... well, depending on how many alternatives are being generated.  If they start to go more than a couple of thousand in one pass, then I start to cut back the score bounds.  Or vice versa, if less than 1k are being generated, I increase the score bounds.

On new tress, or those with less than 500k nodes, I typically use +-5.00 up to +-7.00 depending...depending on whether or not the tree seems to be converging.  By converging, I mean that the tree is generating a likely "good" move.  If it is not converging, then, oddly, I tend to narrow the score bounds... trying to get it to lengthen out variations in the hope of finding the best one.  In general, the narrower the score bounds then the longer the variations analyzed will become. 

Now the ability to judge convergence, well that is a neat trick.  Usually takes experience.  But one trick I use is to judge how long have the top 4 or 5 moves been the top moves?  Without any other moves floating up into the top 5.  If they flip around frequently, then convergence is far away.  And sometimes if the tree is too narrow, a bad move can be converged onto... so judgement is important.
Parent - - By KenB Date 2020-05-31 01:02
Hi Pawnslinger,

Nice to meet you - thanks!  This is very helpful because it gives me some thoughts on how to start using my own judgement in the building of my trees.  I've been worried I'm going to fall into a trap of letting Aquarium do its own thing on autopilot, which I suspect won't take me very far.

There are many options here that seem to narrow/widen the tree in the project properties window.  It's tricky as a novice IDeA user to see how they all interact.  For example, the "Tree shape" drop-down menu gives me a bunch of options to narrow or widen the tree.  Then right below that there's a "Tree width" slider to narrow or widen the tree.  Based on your feedback, the score bounds will be controlling how narrow/wide the tree is.  Of course, the Limit White/Black alternatives will influence this too.  I suspect Preferred Side influences this to some extent as well.  It's not immediately obvious how these all interact to define a "final" tree shape.  I have my own thoughts on when I'd want to narrow or widen the tree, but pulling all the levers to get what I want isn't necessarily intuitive.  I suspect I'll figure this out over time, but I'm starting to wonder if it makes sense to use this Automatic tree expansion properties window sparingly and focus instead on well defined tree generating strategies that give me more granular control over the outcome.

Anyway, I appreciate the insight you provided.  Thanks!
Parent - By pawnslinger (****) Date 2020-05-31 02:13
All I can tell you for sure is... I have been using tree based software for 20 years.  And Aquarium for at least half that time.  Pretty much what I have learned is thru reading this forum and listening to others that use this type of program.  The other main tree software, developed originally in the 90's, is a program now called ChessOpening Wizard.  It used to be called something else, long ago when I first started using it, but darn if I can recall the name.  It has been thru several name changes over the years.  I switched from it to Aquarium, because the other program did not handle long variations or large datasets very well - and is primarily geared toward OTB players.  Anyway, I digress...

I usually set tree shape to Wider and the percentage for tree width to 50 or 60+, depending on how I feel things are going.  If things are going too slow, I try to go narrower, but never less than Moderate 50 percent.  When you look at setting these things, think shape, not really width.  You can have different shapes for the same overall width.  For example you could have a tree shaped like a triangle on its side, kind of like >, or a shape that more resembles a side ways parabola )... but I like to have my tree shaped like a slash /... all could be the same width.  The / tree would have depth of variation determining the shape that is ideal (yet rarely seen).  The goal is to spend one's analyze on likely candidates.

So the score bounds and the tree shape ARE related, but they control different aspects of the tree.
Parent - By Viktor (**) Date 2020-06-01 04:13
Analysis may stop if the root node is not specified. I hope You didn't forget to do this.
Up Topic Rybka Support & Discussion / Aquarium / IDeA running out of tasks?

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill